Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Intel's New Chimera: Alder Lake

agner.org

Post reply on HN
Print view View original

101–110 of 253 posts

Re: Intel's New Chimera: Alder Lake

#101

Post by astrange »

kilburn wrote: post #86

I may be way off here, but to me the transition to heterogeneous cores is more of a long term than short them play, anticipating the era where we'll have different kinds of cores in our systems. Even the performance cores of today have circuitry that's there to make certain workloads faster (video decoding, crypto operations, etc.). Other stuff has mostly been offset to the GPUs (actual grahphics-related stuff, machi…

Intel CPUs already have video encoding/decoding cores built into the iGPU. Though they don’t allocate nearly the space to GPU that they could.

It does provide power advantages more than performance. The downside is that it’s much harder to improve the algorithms after they ship.

Re: Intel's New Chimera: Alder Lake

#102

Post by zasdffaa »

astrange wrote: post #98
deeringc wrote: post #43

Earlier quoted context omitted.

Anyone working on even a medium sized C++ project needs a very high end machine in order to compile. For large projects, 32 core work stations aren't crazy at all. Try building something like Firefox or Chromium on a quad core and you will see what I mean. The cost of the workstation is tiny compared to the cost of the engineer and you really don't want them to be waiting around for half their day (quite literally) b…

Compilation is largely memory bandwidth limited - some of the acceleration from cores might be that they have different memory channels rather than more CPUs. M1 builds a lot faster than Intel CPUs even with fewer cores because of the memory bandwidth.

Really?? I am very surprised, do you have any pointers on this? Kind of important thing to me, thanks.

Re: Intel's New Chimera: Alder Lake

#103

Post by astrange »

roomey wrote: post #33

Just trying to figure this out. So the low power high power cores do seem to be inspired by the M1. If they can implement that easily in x86 is another thing but if anyone can Intel can I guess. The half prescion fps are obviously aimed at ML but how on earth are the hoping to compete with gpus on this front? Where is the use case on this (genuine question). The complex numbers bit makes me think of quantum circuit s…

FP16 is a common GPU format (ML is better with a different one called BF16). The reason to add it to your CPU is so you can work on the same data without having to do software emulation which is really slow.

Re: Intel's New Chimera: Alder Lake

#104

Post by nicoburns »

paulmd wrote: post #69
delusional wrote: post #46

Earlier quoted context omitted.

It's worth making it explicit that many home computing tasks are highly parrallelizable because the work is actually completely separate programs. If you watch a video while writing a document in word, that's trivially parrallelizable.

It’s also not very intensive, especially since video decode usually runs on the GPU. (Or it should - browsers are not always very good about this but the tools are there, they just don’t want to pay licensing fees to use them.) I always cringe a bit when people insist they need 6 cores “so they can have discord and Spotify open while they game”. People don’t generally have a very good intuitive sense of just how litt…

> Playing a video and writing a document is something that can be trivially done on a core2duo era processor, or the modern equivalent of a low-end dual-core nuc with e-cores (atom cores) like a J4005. You don’t even need performance cores for that.

Playing a video and editing a document, sure. But change that to being on a video call and having a few moderately intensive web apps open (say Gmail, Google Docs and some kind of project management software) and your average dual core processor from a few years ago is going to start to struggle.

Re: Intel's New Chimera: Alder Lake

#105

Post by willis936 »

paulmd wrote: post #69
delusional wrote: post #46

Earlier quoted context omitted.

It's worth making it explicit that many home computing tasks are highly parrallelizable because the work is actually completely separate programs. If you watch a video while writing a document in word, that's trivially parrallelizable.

It’s also not very intensive, especially since video decode usually runs on the GPU. (Or it should - browsers are not always very good about this but the tools are there, they just don’t want to pay licensing fees to use them.) I always cringe a bit when people insist they need 6 cores “so they can have discord and Spotify open while they game”. People don’t generally have a very good intuitive sense of just how litt…

Sure chat apps and video playback are light, but what isn't is browsing the web. Web browsing has become a heavy task. If you tried to install a browser and then use it today on a core 2 duo, you'd have a very bad time. And everyone browses the web, usually quite a bit.

Re: Intel's New Chimera: Alder Lake

#106

Post by astrange »

nicolaslem wrote: post #49
zdragnar wrote: post #27

Earlier quoted context omitted.

I thought the secret sauce to M1 was largely that it was a new architecture, without the decades of x86 backwards compatible baggage, with an OS and software capable of running natively on it?

The secret sauce is most likely TSMC 5nm.

No, that’s cope. There isn’t any secret sauce. It’s just well designed on every level.

You won’t match its power/performance by doing one thing and suddenly making it there. Intel’s process is quite close to TSMC’s already.

Re: Intel's New Chimera: Alder Lake

#107

Post by paulmd »

lynguist wrote: post #94
paulmd wrote: post #68

Earlier quoted context omitted.

the concept of “e-cores” is a rebranding of Atom. The Atom name is extremely tarnished from the days when it was super slow in-order netbook cores but Intel started seriously revamping the design in 2012-2014 with silvermont and airmont. They have been good for several years now - Goldmont (eg J5005) is between core2quad and Nehalem i5 (non-SMT) level performance despite running lower clocks (ie IPC is higher), and t…

In my analysis, only Gracemont (Atom 5th gen) is good and that’s for two reasons: 1. It has roughly the IPC of Skylake. 2. It’s the first Atom with AVX2. Even in 2021 Intel released brand new Atom cores without AVX2, which was introduced with Haswell (Haswell New Instructions). And in most tests, merely enabling AVX2 will yield 10% higher performance. That’s why Clear Linux has higher performance as they’re all tuned…

No, it’s really doing OK for what it is. A Xeon Bronze 3104 is a 6C6T 1.7 GHz Skylake that gets a 88W tdp and a J5005 is a 2.8 GHz 4C4T Goldmont running in 15W. That is 10.7 core-cycle-units for the Skylake and 11.2 core-cycle-units for the atom. The Skylake is not really all that far ahead - maybe 50% ahead on average in this suite, so the J5005 is running about 2/3 of Skylake IPC. I assume c-ray is probably using AVX there? And the J5005 still does fine.

There is obviously a dearth of actual practical benchmarks of real tasks apart from STH and a few others, but if you look at Passmark or other generic benchmarks, like I said, it’s certainly faster than core2quad enthusiast desktop, which is completely reasonable and even impressive given its power budget. Actually in Passmark it’s even faster than a full Nehalem Core i7 with SMT, it actually works out just a bit under a Sandy Bridge i5 of similar clock - so I was a bit conservative there. And Passmark corroborates my “3104 is about 50% faster” guesstimate.

(UserBenchmark’s actual measurements - not the speed rating - is likely more accurate than Passmark but I know I’ll get my head bitten off for it!)

https://www.servethehome.com/intel-pentium-silver-j5005-benc...

https://www.cpubenchmark.net/compare/Intel-Pentium-Silver-J5...

(Do remember that Intel progress through that era wasn’t really as bad as people say… I’ve seen people say “5% a generation” and the only generation close to that low was Broadwell. Skylake is a ton faster than Sandy Bridge clock for clock, and J5005 still does fall a bit below Sandy Bridge. There’s your two generations of progress since Goldmont - Atom went from below Sandy Bridge to matching Skylake during the Tremont/Gracemont generations.)

Re: Intel's New Chimera: Alder Lake

#108

Post by moonbug »

xmodem wrote: post #56
em500 wrote: post #37

Earlier quoted context omitted.

The top Alder Lake desktop CPU has 16-cores, 8P+8E, compared to 8P in the 11-th gen. The E cores get around half the performance of the P cores, so the 8P+8E is get roughly the performance of a hypothetical 12P+0E CPU. The very high core parts are reserved for Xeon (a 56 core Alder Lake Xeon has been leaked, but I think the P/E split is still unknown). What people need to realize is that a P cores takes around 4 time…

Indeed, I do sometimes wonder what it'd be like if they made the entire processor out of E cores.

that's the Phi.

Re: Intel's New Chimera: Alder Lake

#109

Post by justsomehnguy »

zdragnar wrote: post #27
quux wrote: post #7

Earlier quoted context omitted.

The goal of this kind of design is M1 like energy efficiency, not performance (I don't know if the Alder Lake P cores can outperform an M1 P core.) M1 has a similar architecture with performance and efficiency cores, but so far I haven't heard of it causing problems for software developers. All versions of macOS that support M1 also have a scheduler that is able to move threads between E and P cores nicely.

I thought the secret sauce to M1 was largely that it was a new architecture, without the decades of x86 backwards compatible baggage, with an OS and software capable of running natively on it?

> that it was a new architecture

This is the biggest part of the 'sauce' but having E-cores and extremely close memory (faster memory access - less time the cores need to wait for data - less time spent doing nothing and burning watts) surely helps.

Re: Intel's New Chimera: Alder Lake

#110

Post by NikolaNovak »

nicoburns wrote: post #104
paulmd wrote: post #69

Earlier quoted context omitted.

It’s also not very intensive, especially since video decode usually runs on the GPU. (Or it should - browsers are not always very good about this but the tools are there, they just don’t want to pay licensing fees to use them.) I always cringe a bit when people insist they need 6 cores “so they can have discord and Spotify open while they game”. People don’t generally have a very good intuitive sense of just how litt…

> Playing a video and writing a document is something that can be trivially done on a core2duo era processor, or the modern equivalent of a low-end dual-core nuc with e-cores (atom cores) like a J4005. You don’t even need performance cores for that. Playing a video and editing a document, sure. But change that to being on a video call and having a few moderately intensive web apps open (say Gmail, Google Docs and som…

I do some work on my ThinkPad t420s. It's about a decade old now. I understand it's anecdotal but fwiw. Note that it's plugged into Ethernet.

I do ALL my work on a t25 which is just about 5 years old now. This includes daily calls with dozens of video participants (ms teams and WebEx mostly) while I have chrome and Firefox open with multiple tabs including outlook and Gmail and slack, couple of MS words, couple of PowerPoints, Visio, etc. That's a dual core processor from more than couple of years ago and it's nowhere near capacity with that workload.

It does have lots of ram and Ethernet to residential Fibe 1gbps. But CPU just hasn't been an issue and I plan to work on this another 5 years

Post reply on HN